Ant Group's artificial intelligence company has launched the LingBot-Depth model.

Ant Group's artificial intelligence company launched the LingBot-Depth model today. According to the introduction, LingBot-Depth is a depth completion model designed for real-world scenarios, aiming to convert incomplete and noise-disturbed depth sensor data into high-quality, true-to-scale three-dimensional measurement results. The model addresses key challenges in robots and autonomous systems: consumer-grade depth cameras are prone to failure on strong reflective surfaces such as glass, mirrors, and metal, which require more reliable depth perception. We treat sensor failure areas as learnable structural clues rather than simple noise, and utilize the visual context information in RGB images through a masked depth modeling mechanism to intelligently infer and complete missing depth areas. At the same time, we have developed an automated data selection and quality control process to support large-scale, scalable training. In order to promote the development of spatial perception research, we will open-source the model, as well as 2 million pairs of high-quality RGB-depth data.